- write stochasitc labs app 🛫 2022-11-15 📅 2023-05-05 ✅ 2023-05-19
ai project proposal
I’m working on Plaintext Transfer Protocol (PTTP), an exploration of what a universal natural language api protocol would look like.
- open-source trend of LLMs… we will soon have comparable (at-cost) models available, what can we do to make the most of them? and how can they help everyday people who are not familiar with these systems
- a lot of the innovation relies on interpreting text. how can we leverage text to empower computers to simply talk to one another without the need for API specifications? Can they force interoperability in a world where platforms are so hostile to that idea? What kind of computer emerges in this world where everything is facilitated by a model? What are the political implications of this?
- we are constantly writing in patterns, how can we leverage LLMs to turn these raw patterns into structured data?
- Whether designating action items to finish with a preceding “TODO” text, timestamping each bullet list item with the date, or “@ing” people, people are constantly creating new ways of patterning that are relatively legible to others. What happens when these syntaxes, created by different people, collide and how is resolution handled between these budding structures?
- I’m working towards prototyping a sample implementation of this protocol, aiming towards an environment where anyone could invent their own social protocols (like spring ‘83 and open stories) and immediately start playing with them in existing apps (e.g. a social protocol around sharing descriptions of the sunrise could semi-automatically syndicate what someone journals in their morning page to a compatible client).
practically, how could open-source LLM models integrate directly into the apps that people use today and empower people (including those non-technical) to create one-off apps to fulfill everyday repeated tasks.
With this kind of tool, you can imagine prototyping and imagining entirely new OS designs revolving around this flexible way of interacting with a computer. What would a journal OS look like where
Programs follow formal syntax and strict format, while human language is not naively interpretable by machines. Even once translated, this data is stuck in siloed apps, leaving no path to using protocols that are not dictated by the host platform. Although new data protocols (Solid, Ceramic, and new chains) point towards a new future where all app data is interoperable, they require you to abandon the old world of apps where everyone already is for the new.
On the internet, social protocols rapidly emerge and die in a vicious evolutionary cycle through people’s digital output. Every digital media object is infused with memetic potential for the invention of new protocol containers, from the structure of a meme for conveying dense amounts of information, to coded language to escape censorship (Chinese citizen net language), to sharing your life story in the comments section (the internet checkpoint). Each of these emergent behaviors and patterns could be distilled into a formal protocol with associated rules for sharing and dictionaries of meanings (and similar phenomena have seen standalone social apps created to capture them).
who do you need in the room for a new kind of internet / computing?
- social media app builders
- activists / organizers who rely on social media
- normal users of social media
- digital media artists
- tech policy makers
- think tanks / institutions / journals thinking about a healthier internet
- research labs ink and switch
- mission-focused VCs / people with money who fund tech things